Read on Twitter
Great thread by @EmilDimanchev here!
Funny thing about path dependency: within certain physical and economic limits, we can *choose* the energy system we want, and then unit costs will come down thanks to learning.
Dorky version: choose a local minima for our non-linear world. https://twitter.com/EmilDimanchev/status/1315619195904720898
Funny thing about path dependency: within certain physical and economic limits, we can *choose* the energy system we want, and then unit costs will come down thanks to learning.
Dorky version: choose a local minima for our non-linear world. https://twitter.com/EmilDimanchev/status/1315619195904720898
There were already well-thought-out concepts for renewable energy systems in the 1970s, but the technologies didn't receive enough support early enough. https://twitter.com/nworbmot/status/1142758825520447488
The Danes pushed wind power down the learning curve in the 80s and 90s (see "Quitting Carbon" by @JustinGerdes), and several countries did the same for PV over several decades (see "How Solar Got Cheap" by @GregNemet), but we could have done it earlier. https://www.howsolargotcheap.com/
I don't doubt that a concerted state-led push for nuclear reactors with wide public support, with a follow-up programme for fast breeders, could have pushed us into a nuclear-dominated equilibrium.
And as @EmilDimanchev points out, we happen to be in crappy fossil-fuelled equilibrium right now, which isn't really optimal on counts that many think are important (air quality, climate, equity, etc.).
And just because a technology is foundering or "non-commercial" right now, doesn't mean that it shouldn't be supported from the multi-decadal perspective of pushing us onto a new pathway of our choice.
There's a nice demonstration of multiple local optima in models with learning in Niclas Mattsson's 1997 licentiate thesis, where he introduced the MILP model of learning curves that was later picked up in several IAMs, see his doctoral thesis summary: https://research.chalmers.se/en/publication/514513
By killing off some branches of the branch-and-bound algorithm using temporary constraints, he could force it down particular paths.
This is a neat trick, since if you're working in a nonlinear system, rather than the piecewise MILP linearisation, it can be hard to find all local minima (e.g. you can try lots of different starting points, simulated annealing etc.).
